Using An Embedded Vision Processor To Build An Efficient Object Recognition System

A brief introduction to the components in a vision processing system and how to to construct one.


Computer vision is a discipline that was established in the 1960s. With the advent of high-performance mobile computing platforms, we see rapid progress in computer vision capabilities. Machine vision is becoming embedded in highly integrated SoCs and expanding into emerging high-volume consumer applications such as home surveillance, games, and automotive safety. A major challenge in enabling mass adoption of embedded vision applications is providing the processing capability at a power and cost point low enough for mobile consumer applications, while maintaining sufficient flexibility to cater to rapidly evolving markets.

This white paper gives a brief introduction to the components of a vision processing system. First, we will discuss the various challenges of efficiently implementing an embedded vision system, taking into account power, performance, memory, as well as software issues. Then we will explore an application example to show how object recognition using the state-of-the-art Convolution Neural Network (CNN) algorithm is done. Finally, we will introduce the DesignWare Embedded Vision Processor Family and describe how an efficient face detection and tracking system that leverages the CNN algorithm is built with one of these processors.

Read this paper to learn about:

  • Applications that use embedded vision
  • Components in a typical vision system
  • Challenges of embedding vision into your SoC
  • Convolutional Neural Networks (CNNs)
  • Architecture and software environment of an efficient embedded vision processor
  • Implementation example of a face tracking application.